Business • Free Strategy Guide

The P.I.T.C.H. Prompting Framework

A safety-first prompting standard for Google Gemini designed to protect pupil privacy and company trade secrets.

Download P.I.T.C.H. Guide (.pdf) Open print-ready prompt list
Free PDF booklet • 600 KB • No registration or email address required
Reference Guide
P.I.T.C.H. Prompting
✓ Persona & Instruction
✓ Tone & Context rules
✓ Redaction protocols

Enable secure inputs

Distinguish between managed Workspace models and unmanaged public tools before pasting records.

Scrub identifiers

Follow the Redaction Protocol to safely query consumer AI models without leaking confidential data.

Standardise outputs

Prevent generic AI drafts by specifying the exact layout, tone, and active instruction structure.

Tier 2 Safe-Start Caveat:

This framework is built to operate safely within a configured, enterprise-grade Google Workspace environment. If your organization relies on standard consumer-grade accounts, you must strictly apply the Redaction Protocol below.

The P.I.T.C.H. Framework Defined

Use this structure as the standard operating procedure for instructing Gemini to perform complex tasks:

P • Persona (Who is the AI?)

Role Definition
Assigning a specific role focuses the AI's internal model parameters and database filters. E.g. "Act as a Senior Data Analyst" or "Act as a KS3 History Teacher".

I • Instruction (What must it do?)

Task Command
The primary command must begin with an active, direct verb. Avoid ambiguous language. E.g. "Summarise this thread", "Draft a lesson plan", or "Analyse these sales figures".

T • Tone (How should it sound?)

Delivery Style
Setting the tone determines the style of the output and prevents typical robotic, generic AI phrasing. E.g. "Professional and concise", "Warm and encouraging (for parents)", or "Strict and factual".

C • Context (The Safety Zone)

Input Parameters
Provide the exact background details required to complete the task. In a Secure Workspace (e.g. Gemini Enterprise), you can paste internal meeting notes or reports. In an unsecure consumer tool, redact names first.

H • How (Format & Layout)

Output Structure
Define the exact layout required so the model does not guess. E.g. "A bulleted list", "A table with columns: Risk, Owner, Mitigation", or "A 200-word email draft".

The Redaction Protocol (For Consumer AI)

If your organisation has not completed a Secure Workspace configuration, staff must use the Redaction Protocol in the Context field before submitting prompts:

✓
Replace specific names with general labels (e.g., [Student A] or [Employee 1]).
✓
Scrub proprietary brand and partner names (e.g., [Client X]).
✓
Mask precise financial targets and budgets (e.g., [Target Budget]).
✓
Generalise geographical or physical coordinates (e.g., [Location Y]).
Standard Operating Procedure

Get the P.I.T.C.H. Prompting Toolkit

Download the complete safety-first briefing and printable quick-reference sheet to distribute across your teams.

Download PDF Guide (600 KB) Open Print-ready Guide